Data-driven event detection for compressed video

A method comprising:

• receiving, from a camera comprising a video encoder, by a first object recognizer deployed in a mobile edge cloud neural network operating at an edge of a mobile telecommunications network, an encoded video comprising a plurality of I-frames, wherein the encoded video comprises a video encoded by the video encoder based upon a best parameter configuration, wherein the best parameter configuration comprises a group of pictures size and a scenecut threshold, and wherein the best parameter configuration is a highest value of a harmonic mean between an accuracy of event detection and a filtering rate;
• processing, by the first object recognizer, at least a first I-frame of the plurality of I-frames in the encoded video;
• and
• after processing at least the first I-frame, providing, by the first object recognizer, at least the first I-frame to a second object recognizer for further processing, wherein the second object recognizer is deployed in a central cloud neural network operating in communication with the mobile telecommunications network and the mobile edge cloud neural network.

A computer-readable storage medium comprising computer-executable instructions that, when executed by a processor of a first object recognizer deployed in a mobile edge cloud neural network operating at an edge of a mobile telecommunications network, cause the processor to perform operations comprising:

• receiving, from a camera comprising a video encoder, an encoded video comprising a plurality of I-frames, wherein the encoded video comprises a video encoded by the video encoder based upon a best parameter configuration, wherein the best parameter configuration comprises a group of pictures size and a scenecut threshold, and wherein the best parameter configuration is a highest value of a harmonic mean between an accuracy of event detection and a filtering rate;
• processing at least a first I-frame of the plurality of I-frames in the encoded video;
• and
• after processing at least the first I-frame, providing, by the first object recognizer, at least the first I-frame to a second object recognizer for further processing, wherein the second object recognizer is deployed in a central cloud neural network operating in communication with the mobile telecommunications network and the mobile edge cloud neural network.